polyaxonmaster-thesis-usHDR

Deep learning HDR DATASET

Bash usage command examples

python ushdrcnn_train.py --epoch=10 --batch-size=15 -g -n -t 
python ushdrcnn_train.py --epochs=10 --batch-size=15 --expo-num=15 -g -n -t

quick test:
python ushdrcnn_train.py --epochs=1 --batch-size=9 --expo-num=15 -g 

Check tensoroard Summary after training

tensorboard --logdir=runs

Polyaxon

Documentation

• Hyper parameter search https://docs.polyaxon.com/concepts/experiment-groups-hyperparameters-optimization/
• Grid search https://docs.polyaxon.com/references/polyaxon-optimization-engine/grid-search/

Config local CLI

polyaxon config set --host=10.23.0.18 --port 31811

If error: polyaxon: command not found 

Try adding the python installation to the PATH
```
python3 -m site &> /dev/null && PATH="$PATH:`python3 -m site --user-base`/bin"
```

Check polyaxonfile.yml

polyaxon check -f polyaxonfile.yaml
upload project:
polyaxon run -f polyaxonfile.yaml -u -l

where

• -u: upload/update project
• -f update polyaxonfile.yaml
• -l enable logs onto the standard output (the terminal)

Check experiment:

http://cluster.ifl:31811

Datasets	Inputs	Input_imgs
newDataset	1901	28515
LDR2_fakeComp_DataSet	26	390

new_DataSet

File Structure

├──Originals/
|  ├─Study/
|  ├─ date-generic/
|     |- 00.b8 -> 15.b8
├──images/
|  ├──Study/
|        | exVivo_xx.png
|        | exVivo_power.txt
|        ├─Results/
|           | - stack_hdr_image.hdr
|           | - hdrReinhard_global.png
|           | - hdrReinhard_local.png

LDR_DataSet

Folder Structure

├──Org_images/
|   ├── LDR_01469_0xxxx.tiff
├──c_images/
|   ├──LDR_0xxxx/
|       | exVivo_xx.png
|       | exVivo_power.txt
|       ├──Results/
|           | stack_hdr_image.hdr
|           | hdrReinhard_global.png
|           | hdrReinhard_local.png

Fakehdr dataset v2

biggest image: LDR_01469.tiff

maxH: 839	maxH: 632

Smallest image: LDR_00864.tiff

minH: 236	minH: 279

File Structure

├──Org_images/
|   ├── LDR_0xxxx.tiff
├──c_images/
|   ├── LDR_0xxxx/
|        | exVivo_xx.png
|        | exVivo_power.txt
|        ├──Results/
|           | stack_hdr_image.hdr
|           | hdrReinhard_global.png
|           | hdrReinhard_local.png

	| Training SETUP:       | Epoch finished ! 						|
	|Epochs: 1		| Train Loss:0.000266						|
	|Batch size: 15		| Val Loss:0.000046, running_val_loss:0.000046  		|
	|Learning rate: 0.1	| Validation loss: 0.000046188					|
	|Training size: 3042    | Training complete in 68m 48s					|
	| Validation size: 760  |								|
	|Checkpoints: False	|								|
	|CUDA: True		|								|	
	| ----------------------|---------------------------------------------------------------|



	| Training SETUP:       | Epoch finished ! 						|
	|Epochs: 8		| Train Loss:0.000274						|
	|Batch size: 15		| Val Loss:0.000049, running_val_loss:0.000049  		|
	|Learning rate: 0.1     | Validation loss: 0.000049238					|
	|Training size: 3042    | Training complete in 399m 26s					|
	| Validation size: 760  |								|
	|Checkpoints: False	|								|
	|CUDA: True		|								|	
	| ----------------------|---------------------------------------------------------------|

Interesting links:

https://machinelearningmastery.com/learning-curves-for-diagnosing-machine-learning-model-performance/